[{"data":1,"prerenderedAt":220},["ShallowReactive",2],{"topic-page:\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications\u002Ffoundation-model-selection-and-inference-parameters":3},{"topic":4,"category":175,"seo":179,"breadcrumbs":184,"prerequisites":194,"nextTopics":195,"relatedTopics":204},{"id":5,"slug":6,"title":7,"summary":8,"canonicalPath":9,"categoryPath":10,"difficulty":11,"targetAudience":12,"estimatedMinutes":13,"generatedAt":14,"publishedAt":14,"updatedAt":14,"reviewStatus":15,"tags":16,"expectedKnowledge":25,"learningGoal":28,"keyTerms":29,"blocks":87},"it-aws-aws-certified-ai-practitioner-foundation-model-applications-foundation-model-selection-and-inference-parameters","foundation-model-selection-and-inference-parameters","基盤モデル選択と推論パラメータ","基盤モデル選択と推論パラメータを、AIF-C01の公式objective 3.1.1・3.1.2に沿って、用語、判断順序、例、誤解修正で整理します。","\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications\u002Ffoundation-model-selection-and-inference-parameters","\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications","beginner","cert_candidate",6,"2026-08-06","ai_generated",[17,18,19,20,21,22,23,24],"AWS Certified AI Practitioner","AIF-C01","基盤モデルの活用","基盤モデル選択","temperature","入力長","出力長","プロンプトキャッシュ",[26,27],"AWSクラウドの基本用語を聞いたことがある","AIを使う業務例を日常的な言葉で考えられる","基盤モデル選択と推論パラメータについて、試験で問われる違いと選択基準を説明し、短い業務例へ適用できる。",[30,38,44,48,52,55,58,65,69,73,77,83],{"id":31,"label":32,"shortDefinition":33,"supplement":34,"role":37},"inference-parameters","推論パラメータ","学習済みモデルを変えず、生成時の候補選択や出力量を制御する設定です。",{"kind":35,"text":36},"misconception","設定変更だけでは、モデルの能力や事実性は追加されません。","primary",{"id":39,"label":40,"shortDefinition":41,"supplement":42,"role":37},"model-selection","モデル選択","用途の能力・制約・費用・統制条件に合うモデルを比較して選ぶ活動です。",{"kind":35,"text":43},"最大のモデルが、すべての用途で最適とは限りません。",{"id":21,"label":21,"shortDefinition":45,"supplement":46,"role":37},"生成候補の選び方の多様性を調整する推論パラメータです。",{"kind":35,"text":47},"低くしても、回答の正しさが自動で保証されるわけではありません。",{"id":49,"label":50,"shortDefinition":51,"role":37},"context-window","コンテキストウィンドウ","モデルが一回の処理で参照できるトークン範囲です。",{"id":53,"label":23,"shortDefinition":54,"role":37},"output-length","一回の生成でモデルが返せる出力の長さです。",{"id":56,"label":24,"shortDefinition":57,"role":37},"prompt-caching","繰り返す入力部分の処理結果を再利用し、遅延や費用を抑える仕組みです。",{"id":59,"label":60,"shortDefinition":61,"supplement":62,"role":64},"foundation-model","基盤モデル","大規模なデータで事前学習され、多様な下流タスクへ適応できるモデルです。",{"kind":35,"text":63},"すべての定型予測を、基盤モデルへ置き換える必要はありません。","contextual",{"id":66,"label":67,"shortDefinition":68,"role":64},"modality","モダリティ","文章・画像・音声・動画など、モデルが扱うデータ形式の種類です。",{"id":70,"label":71,"shortDefinition":72,"role":64},"latency","レイテンシ","要求を送ってから結果が返るまでにかかる時間です。",{"id":74,"label":75,"shortDefinition":76,"role":64},"model-complexity","モデルの複雑さ","モデルの規模・構造・計算量など、運用負担へ影響する複雑さです。",{"id":78,"label":79,"shortDefinition":80,"supplement":81,"role":64},"retrieval-augmented-generation","検索拡張生成（RAG）","関連情報を検索し、その結果を文脈に加えて回答を生成する方式です。",{"kind":35,"text":82},"検索結果が正しいか、回答の根拠確認は必要です。",{"id":84,"label":85,"shortDefinition":86,"role":64},"evaluation","評価","定めたデータと基準で、モデルやアプリの品質・安全性・価値を測る活動です。",[88,93,104,113,122,128,134,140,156,165],{"id":89,"type":90,"title":7,"subtitle":91,"shortDefinition":92},"hero","HeroBlock","モデル能力と、温度・入出力長の役割を分けて調整する","学習済みモデルそのものを変えず、生成時の候補選択や出力量などを制御する設定です。",{"id":94,"type":95,"term":32,"definition":96,"plainExplanation":97,"presentation":98,"supportLabel":99,"keywords":100},"definition","DefinitionBlock","[[cloze:inference-parameters|推論パラメータ]]とは、学習済みモデルを変えず、生成時の候補選択や出力量を制御する設定です。","[[term:model-selection|モデル選択]]では、[[term:foundation-model|基盤モデル]]の能力、費用、[[term:model-complexity|モデルの複雑さ]]を比較します。選択後は[[term:temperature|temperature]]で候補の多様性を、[[term:context-window|入力長]]と[[term:output-length|出力長]]で扱う情報量を調整します。[[term:prompt-caching|プロンプトキャッシュ]]は繰り返す入力の処理を再利用します。","natural","見分けるポイント",[101,102,103],"候補選択","文脈範囲","生成量",{"id":105,"type":106,"steps":107,"title":112},"decision-steps","StepBlock",[108,109,110,111],"用途の必須[[term:modality|モダリティ]]・言語・文脈長・統制条件でモデルを選ぶ。","代表入力で品質、[[term:latency|遅延]]、費用を比較する。","創造性と再現性の要求に合わせてtemperature等を調整する。","入出力長、停止条件、キャッシュ効果を測り、全条件を[[term:evaluation|評価]]する。","選択と調整を分ける",{"id":114,"type":115,"title":116,"points":117,"body":121},"exam-cues","KeyPointBlock","判断で押さえる境界",[118,119,120],"必須能力はモデル選択で満たす","推論パラメータは生成時の挙動を調整","temperature低下は事実保証ではない","モデルを替える問題と、同じモデルの生成条件を調整する問題を混同しません。画像入力に非対応ならtemperatureでは解決できず、モデル選択へ戻ります。回答が冗長なら出力上限や指示を見直せますが、短くし過ぎると必要事項を欠きます。パラメータごとに変えられる性質を確認します。",{"id":123,"type":124,"src":125,"alt":126,"caption":127},"diagram","DiagramBlock","\u002Fassets\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications\u002Ffoundation-model-selection-and-inference-parameters\u002Ffoundation-model-selection-and-inference-parameters.svg","基盤モデル選択と推論パラメータについて用途要件、FM能力・制約、温度・入出力長、評価：品質・費用・遅延の関係を文字ラベルと矢印で示した図","変えにくいモデル能力と、一回ごとに調整する推論設定を層で分けた図です。最後に共通評価へ戻します。",{"id":129,"type":130,"scenario":131,"explanation":132,"result":133},"worked-example","ExampleBlock","規程FAQでは回答を簡潔かつ安定させ、広告案では多様な候補を出したい状況です。","両方で同じFMが最低要件を満たすなら、FAQは低めのtemperatureと短い出力形式、広告案は評価済みの範囲で多様性を上げる設定を試します。ただしFAQの根拠不足はtemperatureでは直らないため、[[term:retrieval-augmented-generation|RAG]]や資料品質を見直します。設定変更ごとに代表質問で誤りと費用を測ります。","モデル能力、文脈、推論設定のどこを変えるべきか診断できます。",{"id":35,"type":135,"warningTitle":136,"message":137,"severity":138,"action":139},"WarningBlock","temperatureを0にしても真実保証にはならない","出力が安定しやすくなっても、モデルが持つ誤情報や不足した文脈は残ります。正確さには根拠、検証、評価が必要です。","warning","要件違反が能力不足、文脈不足、生成条件のどれかを切り分けます。",{"id":141,"type":142,"title":143,"questions":144},"quiz","QuizBlock","AIF-C01 判断チェック",[145],{"question":146,"choices":147,"answerIndex":151,"choiceExplanations":152},"画像入力が必須なのに候補モデルが文章専用の場合、最初に行う判断はどれですか。",[148,149,150],"画像対応モデルへ選択基準を戻す","temperatureだけを上げる","出力上限だけを短くする",0,[153,154,155],"正解です。必須モダリティはモデル能力です。","入力能力は追加されません。","入力モダリティの不足は解消しません。",{"id":157,"type":158,"title":159,"points":160},"summary","SummaryBlock","まとめ",[161,162,163,164],"FMは能力・費用・遅延・長さ・統制で選ぶ","推論パラメータは生成時挙動を調整","設定ごとに代表入力で評価する","能力・文脈・設定の問題を切り分ける",{"id":166,"type":167,"title":168,"nextTopics":169,"relatedKeywords":171},"next-action","NextActionBlock","次は「RAGとベクトルデータベース」へ",[170],"it-aws-aws-certified-ai-practitioner-foundation-model-applications-rag-and-vector-databases",[172,173,174],"RAG","埋め込み","ベクトルデータベース",{"path":10,"title":19,"description":176,"parentPath":177,"accentToken":178},"基盤モデルの選択、RAG、プロンプト、カスタマイズ、エージェント、評価をAIF-C01の設計判断として整理します。","\u002Fit\u002Faws\u002Faws-certified-ai-practitioner","cloud-amber",{"title":180,"description":181,"canonicalUrl":182,"ogImage":183},"基盤モデル選択と推論パラメータ | AWS Certified AI Practitioner","AIF-C01対策として基盤モデル選択と推論パラメータの定義、判断軸、具体例、誤解しやすい境界を公式試験範囲に沿って学びます。","https:\u002F\u002Fzeqnilo.com\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications\u002Ffoundation-model-selection-and-inference-parameters","https:\u002F\u002Fzeqnilo.com\u002Fassets\u002Fsite\u002Fdefault-og.svg",[185,188,191,192,193],{"title":186,"path":187},"IT","\u002Fit",{"title":189,"path":190},"AWS","\u002Fit\u002Faws",{"title":17,"path":177},{"title":19,"path":10},{"title":7,"path":9},[],[196],{"id":170,"title":197,"summary":198,"canonicalPath":199,"categoryPath":10,"categoryTitle":19,"difficulty":11,"targetAudience":12,"estimatedMinutes":13,"publishedAt":14,"updatedAt":14,"reviewStatus":15,"tags":200},"RAGとベクトルデータベース","RAGとベクトルデータベースを、AIF-C01の公式objective 3.1.3・3.1.4に沿って、用語、判断順序、例、誤解修正で整理します。","\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Ffoundation-model-applications\u002Frag-and-vector-databases",[17,18,19,172,173,174,201,202,203],"Amazon Bedrock Knowledge Bases","OpenSearch","Aurora",[205],{"id":206,"title":207,"summary":208,"canonicalPath":209,"categoryPath":210,"categoryTitle":211,"difficulty":11,"targetAudience":12,"estimatedMinutes":13,"publishedAt":14,"updatedAt":14,"reviewStatus":15,"tags":212},"it-aws-aws-certified-ai-practitioner-generative-ai-fundamentals-aws-genai-services-infrastructure-and-cost-tradeoffs","AWS生成AIサービス・基盤・費用トレードオフ","AWS生成AIサービス・基盤・費用トレードオフを、AIF-C01の公式objective 2.3.1・2.3.2・2.3.3・2.3.4に沿って、用語、判断順序、例、誤解修正で整理します。","\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Fgenerative-ai-fundamentals\u002Faws-genai-services-infrastructure-and-cost-tradeoffs","\u002Fit\u002Faws\u002Faws-certified-ai-practitioner\u002Fgenerative-ai-fundamentals","生成AIの基礎",[17,18,211,213,214,215,216,217,218,219],"Amazon Bedrock","SageMaker AI","Amazon Q","Kiro","Strands Agents","Bedrock AgentCore","生成AI費用",1787370539879]